Help

What is AI Music Seed?

AI Music Seed turns a chord progression you type into a short audio file — plus a matching MIDI — that you hand to an AI music generator like Suno so it follows the harmony you chose instead of inventing its own.

Here's the catch it solves: Suno and tools like it don't read chord names from your text prompt. They only follow harmony they actually hear. Writing "Am C G Dm" in the prompt does nothing; the generator never sees it as chords. So other tools that give you chord names — or a MIDI Suno can't listen to — don't help. AI Music Seed renders your progression to real audio. You upload that audio as a Cover or Extend seed, and the generator is locked to your chords.

The core of the app is one action: write a progression and press Render seed — everything else is a dial on that. The window has two tabs: Music holds the music itself — the chords, the figure the two players make of them, and a band of sampled instruments with a bass and a treble player you pick and mix independently — while Render holds everything about the output file: the room, the filters, the format. The Music tab is also a step editor: you can pick a ready-made style, or write the figure yourself note by note and hear it change under your hands. You can hear it all live with Play before rendering anything, editing the music while it plays. A seed is kept as a small .yams document you can name, reopen, and keep several of open in tabs; you save one before rendering, and the audio lands right beside it. This help walks through generating a seed, then how to save and manage your seed files, and finally language and appearance.

Generate a harmonic seed

This is the heart of the app. You type a chord progression, choose a style and the instruments that play it, and press Render seed. The app writes two files next to your saved seed — the audio seed and a matching MIDI — and shows you their paths so you can open or upload them. (Saving the seed itself as a .yams document comes first, and is covered in a later section — the render lands in that file's folder.) The rest of this section explains each thing you can set.

The render button itself, Render seed, lives on the Render tab and runs the generation. It's disabled while a render is in progress, while the form has nothing to render or a problem the preview has flagged (see See the seed before you render, below), and until you've saved the seed to a .yams — because that file's folder is where the audio is written. When it's ready, press it and the files are saved.

What a harmonic seed is

A harmonic seed is a few seconds of audio that plays your chord progression — nothing more. It's the "seed" you give an AI generator to grow from.

It exists because of how generators like Suno work. They ignore chord names in the text prompt entirely and shape their output around the harmony in the audio you upload. The seed is simply how you hand them that harmony in a form they can hear. Reach for one whenever you want the AI to stay on a specific progression rather than make up its own.

Writing the chords

The chords are the one thing you must type. They live in the Chords row at the bottom of the Music tab, which shows them as blocks laid over the beats they occupy; press Edit the chords as text — Ctrl+Enter or Escape to come back to swap those blocks for the text and type into it, and Ctrl+Enter or Escape to come back. Write chord symbols separated by spaces. How the lines you type map to time is covered next, in Lines are bars; this part is just the notation for a single chord.

The notation covers most common chords:

Triads — C, Am, F, G. A bare letter is major; add m for minor.
Sevenths and extensions — G7, Cmaj7, Am7, D9, Em11. Write the quality straight after the root.
Suspended and altered — Csus2, Dsus4, Bdim, Faug.
Slash bass — put a bass note after a slash: D/E, Bm/F#, Gmaj7/B. The chord stays the same; the note after the slash is played in the bass.
Rests — N.C. ("no chord") leaves a silent gap. You can still give it a bass note: N.C./B plays just that bass tone.

The sharp or flat goes right after the note letter (F#, Bb). A quality the app doesn't know written on a real note — a stray suffix, say — falls back to a plain major triad on that root rather than failing; but a symbol it can't read as a chord at all is flagged as unrecognised in the live preview, and the render stays disabled until you fix it.

A chord can also carry a weight. Put a number in front of it and it lasts that many of the line's own units: 2Am 2E G B is six units on that line, so Am and E each hold twice as long as G and B. Without any numbers a line simply splits evenly between its chords, exactly as before.

This is not the same as writing the chord twice, and the difference is audible. Every chord speaks on its own beginning, so Am Am strikes A minor a second time; 2Am lets the first strike ring on while the figure keeps running over it. On a sustained pad that extra attack re-articulates the chord; under a busy arpeggio it barely shows. Reach for a weight when a chart holds a chord across uneven beats — the pattern you can't write with an even split.

Example. Am is A minor, Cmaj7 is a C major-seventh, and Am F C G/B is a four-chord run that puts B in the bass under the final G. A bare G and Gmaj are the same chord; G7 adds the flat seventh. And 2C G Am gives C half the line with G and Am sharing the rest.

Lines are bars

The chord text reads one line as one bar. Every non-blank line you type is a single bar of time, and the chords on that line share it — equally by default, or in proportion to their weights if you wrote any. Empty lines, [Section] labels and | bar marks are ignored, so you can paste a chord chart in almost exactly as it's written and it lines up.

This is what lets a real chart work without reformatting. A line like [Verse] disappears, a blank line between sections disappears, and the | you use to separate bars is just stripped — only the chords and their line breaks decide the timing.

How long a line lasts is itself adjustable, with the + / − buttons on the right of the chords, beside the progression: Longer lines: each line of chords lasts a quarter-bar more and Shorter lines: each line of chords lasts a quarter-bar less. The number under them is the current setting, and it moves a quarter-bar at a time. Normally leave it at one — a line is a bar. Raise it to stretch every line out, or drop it below one to compress them. It belongs to the chords rather than to the seed as a whole, because it says nothing about the music except how the text you typed is read.

Example. Typed on four separate lines, Am, C, G, Dm are four bars, each chord held a full bar. Typed as a single line Am C G Dm, they're one bar with each chord squeezed into a quarter of it. And C G Am F on one line is one bar of four quick chords, while the same four on four lines is a slow four-bar loop.

Tempo, length and meter

A short row of settings above the music decides how fast the seed runs, how a bar is divided, and how long the whole thing lasts. The AI picks up tempo and feel from the seed, and the seed needs to be long enough to give the generator something to hold onto.

BPM — the tempo, in beats per minute. This is the speed the generator tends to carry into its result.
Time signature — the time signature. It's a menu rather than a text field, listing the meters the app actually plays, split into Simple (4/4, 3/4, 2/4 and the like) and Complex (5/4, 7/8 and their relatives). A seed written in a meter that isn't on the list still opens and still plays with the value it was saved with.
Beat division — how finely a beat is divided, and so how many steps a bar of the editor holds. B/2 is eighth notes, B/4 sixteenths, B/3 and B/6 the triplet grids. It applies to both players at once, because they share one time axis. Changing it re-times the figure rather than refusing: switch a run of eighths to B/3 and it becomes the same run played as triplets.
Swing — how much the off-beats are pushed late, as four named amounts: None, Light, Shuffle and Hard. Straight time is None; Shuffle is the classic triplet feel.
Loops — how many times the whole progression repeats in the rendered file. More loops make a longer seed. (Live playback simply loops until you stop it, so this setting only shapes the file.)

Example. At 80 BPM in 4/4, a line holding one chord lasts one bar — three seconds. Four such lines make a twelve-second pass; with 4 loops the seed is about 48 seconds long. Halve the BPM to 40 and every duration doubles.

Styles and figures — how the chords are played

The harmony is the same however you play it; what changes is the figure the two players make of it. That matters because the generator absorbs the seed's texture along with its harmony — the choice trades a clear harmonic bed against how much of the seed's own character leaks into the result.

There are two levels of choice, and they sit in different places.

A style is a named pairing: one figure for the chords and one for the bass, plus the settings that pairing wants. It's the quick choice — the Style menu near the top of the Music tab, sitting directly above the two players it pairs. The styles come in three collections, grouped by what kind of figure they play:

Pads — styles that state the harmony without a moving figure. Pad holds the whole chord as a sustained bed; Drone is the same with the root doubled an octave lower — a deeper anchor, slightly softer; Marker plays one short stab at each chord change and then silence — the closest thing to a silent skeleton that nudges the AI without colouring it.
Arpeggios — broken-chord figures over a simple bass, from a plain up-and-down (Arpeggio 1) through picked eighth-note lines to a two-bar call-and-response and a figure that deliberately cycles out of step with the bar and realigns every three bars.
Grooves — rhythm-forward styles built on a moving bass line with pushes and passing notes under sustained chord stabs.

A figure is one player's own steps, and each player has its own menu at the start of its row — the treble picks from a collection of chord figures, the bass from a collection of bass figures. They are separate libraries, deliberately kept short and plain: named by the order of the chord tones they walk (1-3-2, 1-2-3-4), plus the two that don't walk at all — the whole chord, and a single stab.

Mix the two freely. Take a style you like and swap only its bass figure; the sound is yours, and the style menu changes to Custom to say so — the pairing is no longer the preset's. Picking any style again resets both players to it, which is always the way back.

If you're unsure, Pad is the safe default: present enough to lock the harmony, neutral enough to stay out of the way. The best way to choose is to press Play and step through the menu while it plays — the style changes on the next chord, live.

Writing your own figure — the step editor

When no shipped figure does what you want, write your own. The Music tab isn't a form describing the music, it is the music: a grid you draw the figure into, playing back as you edit.

Three blocks are stacked there, sharing one time axis: the treble player's grid on top, the bass player's below it, and the chords underneath. A column is a fixed slice of time — one step of the beat division — and it means the same instant in all three, so a chord sits directly over the beats it owns and the two players line up with each other and with it. The grid is two bars wide and grows a bar at a time when a player or a progression line needs more room.

Each player draws only its own steps. A six-step figure under an eight-step one is six cells and then grey: past a player's own length the row is dimmed and can't be clicked, because there is nothing there to edit. The players are not locked to the same length — they run at their own, side by side, and a figure whose length doesn't divide the bar simply carries on across it.

The rows are chord degrees, not pitches. That is the idea that makes one figure play a whole progression, and it's worth a moment. Reading a grid from the bottom up:

One chord tone: 1 = root, 2, 3, 4 upward; past the chord's size, the tone an octave up (Chord Player's C1–C4). Stack several in one column for a chord of your own — the numbered rows are chord tones. Row 1 is the root of whatever chord is sounding at that instant, 2 the next tone up, and so on. So a figure written as 1-3-2 plays A, E, C over an Am and C, G, E over a C — the shape stays, the notes follow the harmony. A degree past the chord's size (row 4 on a three-note chord) is that tone an octave higher.
The fifth above the root — Chord Player's BF row — the f row is the perfect fifth above the root, whatever the chord's own tones are. Useful under a bass line.
The whole chord in this step — Chord Player's tall block (three notes on a triad, four on a seventh). Click a tone row in its column to start the block there — the r row places the whole remaining chord in one step: three notes on a triad, four on a seventh. It draws as a dot on the r row with a block below it covering the tones it holds, so you can see at a glance what it is sounding. Click a numbered row inside or under that block to move where the block starts — r from row 3 upward leaves rows 1 and 2 free for something else.

Stack several notes in one column and they sound together — that's how you write a voicing of your own rather than using r.

While Play is running, the column being played lights up in each block, and the sounding chord lights with it. Each player's playhead runs at its own pace, because each runs at its own length — that's not a glitch, it's the two figures cycling against each other.

Anything you draw makes the style read Custom, and the figure becomes the seed's own. It is saved inside the .yams in full, so a seed plays the same in a year even if the shipped figure you started from is later improved. Picking a style from the menu drops your edit and goes back to that preset — including re-picking the one you started from, which is how you undo an experiment.

Placing notes, and the two click modes

A click on the grid can mean two different things, and a small mode switch at the top of the flags column decides which. This is the one thing worth learning before anything else in the editor.

Edit notes: click a cell to add a note, click a note to remove it — the drawing mode, and where you'll spend most of your time. Click an empty cell to put a note there; click a note to take it away. A note you place is also selected, so you can flag it straight away.
Note settings: click a note to select it, then set its flags — the adjusting mode. A click selects the note under it and never writes, so you can reach a note you drew earlier and change its flags without deleting it on the way in.

Both modes exist because a single click cannot carry both jobs. In drawing mode the only click that could select an existing note is the one that removes it; without the switch, adjusting a note would mean switching it off and on again — which rewrites the note you were trying to adjust and loses every flag it carried.

There is a shortcut for the common case: Ctrl+click (or a right-click) selects a note without leaving drawing mode. Use the switch when you're going over a figure you already wrote; use Ctrl+click when you're drawing and want to flag something in passing.

If your player is set to a coarser division than the seed's — a figure written in eighths while the grid shows sixteenths — clicking a column between two of its steps refines that player to the finer grid, keeping everything it already played where it was, and puts the note on the column you clicked. You never have to think about it; the figure simply gains the resolution it needs.

Note flags — shaping a single step

The column of buttons to the left of the grids acts on the selected note — the one with a ring around it. With nothing selected they're greyed, because there is nothing to flag. Each pair is one property:

octave — Octave up and Octave down move just that note a whole octave, leaving the rest of the figure where it is. This is how a bass line reaches down for one step, or an arpeggio tops out.
scale step — Passing note: one scale step up and Passing note: one scale step down turn the note into a passing note, one step off the chord tone. The step is taken through the scale your progression actually spells out, not a chromatic semitone, so it stays inside the key without you having to name one.
level — Accent plays the note harder than the rest of the player's line, Ghost much softer. A figure with one accent and a couple of ghosts reads as played rather than typed.
length — Sustain to the end of the chord rings the note until the chord changes, whatever the player's normal note length; Staccato cuts it short. These are the two ends, and everything between them is the player's own length setting.
condition — Only in the last beat of the chord and Only outside the last beat make the note conditional on where the chord is: one fires only in the chord's final beat, the other only outside it. That's how a figure gets a turnaround — an extra note that appears just before each chord change and nowhere else.
push — Early — this slot voices the next chord voices the next chord in this step instead of the current one, so the harmony arrives a beat early. It's the anticipation that gives a groove its lean, and it wraps around properly at the end of the loop.

Flags stack: a note can be an accented staccato passing note that only fires in the last beat. The grid shows them as small marks on the cell, so a glance tells you which steps carry something.

Each player's own settings

Two columns of small buttons sit on the right of each grid, row for row with it, and they act on that player as a whole rather than on one note:

Add a step at the end of the lane and Remove the lane's last step change the figure's length in steps — the number below them is the current count (Steps in the lane — the + and − above add or remove one). Add steps and the figure gets longer, which is how a two-bar phrase or a deliberately odd length gets written; the grid grows a bar when it needs to.
Shift left and Shift right rotate the whole figure by one step, which is usually faster than redrawing a phrase that starts on the wrong beat.
Play the lane back to front. A setting, not an edit — the figure keeps its name and it toggles off again — plays the figure back to front. It is a setting, not a rewrite: the grid shows it reversed, the figure keeps its name, and pressing it again puts it back. An up-and-down arpeggio mirrored is the same figure descending.
Clear the lane — empties the player, for starting again.

On the band above each grid, beside the instrument, sit that player's note settings — the defaults every step inherits unless a flag overrides it:

velocity — how hard the player strikes, as a percentage. The accent and ghost flags scale up and down from this, so it sets the line's overall energy.
length — how long a note rings, counted in steps. Longer than one step means notes overlap, which is what stops a figure sounding like separate blips; shorter makes it crisp.
Hold — ring every note until the chord changes, regardless of the length setting. This is what turns a figure into a pad; it also switches the player to a sustained voice rather than a struck one. Bass lines usually want it on, arpeggios off.

Instruments — a bass and a treble player

The seed is played by up to two instruments at once, one per register: a bass player carrying the low line and a treble player carrying the chords above it. Each has its own row on the Music tab, labelled Bass and Treble at the end of the row. The instruments come from a built-in library of nearly a hundred sampled acoustic, electric and synthesised instruments — keys, plucked and bowed strings, winds, tuned percussion, voices and synth pads — recorded from openly-licensed sample sessions, plus a clean built-in synth.

Each picker is a pair: a family icon and the instrument list. Click the icon to choose a family — clicking one immediately switches the player to that family's first instrument, so the sound follows the click; choose All to browse everything. The list shows each instrument once; typing in it filters by name (Type to filter…). Two chevrons beside the name step straight to the Previous instrument in the list or Next instrument in the list without opening the list — the quickest way to audition neighbours while something is playing.

When an instrument was recorded in several articulations — sustained, staccato, pizzicato, vibrato — every one of them is a small button on that player's row. They're all on screen at once rather than hidden in a list, so switching is a single click; the button shows a short form of the name and its full name is in the tooltip.

The bass picker offers one extra choice at the top: Same as instrument, which means the treble instrument covers both registers — one player for the whole seed, the way a solo piano would play it. Instruments that only make sense in the low register don't appear in the treble list.

The mixer — each player's knobs

Further along each player's row sits its mixing channel — a mute button and a set of knobs. Drag a knob up or down to turn it, or click its value and type; the ones that measure an amount draw a filled arc from their minimum, so a glance across the row tells you where everything is set.

The mute button silences that player without removing it — the other keeps playing. It's a listening tool: mute the bass to judge the treble on its own, flip it back mid-play. Muting is never saved into the seed, so a reopened seed always plays complete. (A render made with a player muted gets a -bass or -treble suffix on its file name, so a partial file can't be mistaken for the full arrangement.)
volume — that player's level, as a percentage. 100% is untouched.
octave — shifts that player's whole part up or down by one octave. Some instruments simply sit better an octave away from where the figure writes them.
reverb — how much of that player feeds the room chosen on the Render tab (see The room and the master filters). There is one shared room; this knob is each player's send into it. New seeds start the bass low — low frequencies drowned in reverb are where muddiness comes from — and the treble at the room's natural level.
tune (treble only) — a fine-tuning probe in cents, for checking an instrument against another by ear. It resets when you switch instruments and is never saved; leave it at 0 in normal use.

Hear it live — Play and the keyboard

Play (on the Render tab, beside the filters) plays the seed immediately, looping until you press Stop — no render, no file, no saving required. It exists so you can design the seed by ear: while it plays, every musical setting stays editable — the chords, the style, both instruments, the tempo, the meter, the mixer knobs — and each change is heard on the next chord, without stopping. Step through styles, swap instruments, pull the bass down an octave, and only render when it sounds right.

Below the music controls, a piano keyboard shows what's sounding: the treble player's notes light in one colour and the bass player's in another, live. You can also click any key to hear that single note dry, played by the current treble instrument — handy for checking a sound before committing to it.

Playback is a preview of the same arrangement the render writes — same notes, same instruments, same mix. Two small differences are deliberate: the file renders exactly the number of loops you set (playback just loops on), and the master filters are read once when Play starts, so changing those two needs a stop and start.

The room and the master filters

Three settings on the Render tab shape the sound of the finished file, after the instruments have played:

Reverb — the room the seed is placed in, from Dry: no room through Room: small and close, Chamber: medium, warm, Hall: large and open and Church: very large, long tail. One room for the whole seed; how much of each player is sent into it is that player's reverb knob on the Music tab. Choose Dry: no room for a completely dry seed.
Highpass (Hz) — removes everything below the given frequency, in hertz, at the very end of the chain. Below the music sits rumble that eats headroom without being heard as music. 20–80 Hz is always safe; the default 100 already grazes the lowest bass notes, and values above ~150 audibly thin the bass. If you drop the bass an octave, keep it at 60 or below. 0 switches it off.
Lowpass (Hz) — the same at the top: removes everything above the given frequency, where sample noise and compression fizz live. 8,000–16,000 Hz is always safe; the default 10,000 is transparent; lower values make the seed progressively warmer, then muffled. 0 switches it off.

The three exist because a seed should hand the generator music, not noise: the room gives the harmony a natural space, and the filters trim the extremes so the file's headroom is spent on the chords.

Audio format

The Format setting, on the Render tab, picks how the audio seed is written. A seed you upload to an AI generator doesn't need studio fidelity, so the default keeps the file small.

MP3 — compressed and small; the right choice for almost every upload.
WAV (uncompressed) — the raw audio, much larger. Use it only if you specifically want an uncompressed file.

When the format is MP3, MP3 bitrate sets the quality — higher numbers mean a slightly larger, slightly cleaner file. For a Suno seed, 192 kbps is plenty. The bitrate control is greyed out when WAV is selected, because WAV has no bitrate to set.

Naming the seed

Name, on the Render tab beside the other output settings, is what this seed is called — "Chorus lift", "Two Names, One Face", whatever tells you which idea it holds. It's a label for you, not a setting that changes the audio, and it's saved inside the seed so it comes back when you reopen it.

It has one practical job beyond that: it feeds the name the rendered files get, so naming a seed is usually enough to name everything it produces. Leaving it empty is fine — the files then fall back to a name built from your chords (see Naming the rendered files).

Naming the rendered files

The seed has a name; the files it renders have their own. Output is that second name — the base shared by the audio and the MIDI — and it is written as a small template rather than a fixed word, so each render names itself from the settings you were using.

Anything in curly braces is filled in when you render. The musical tokens: {name} becomes the seed's name, {chords} a name built from the chords (the first eight, so a long chart doesn't produce an unusable file name), {style}, {bpm} and {loops} the current values of those settings, and {instrument-treble} / {instrument-bass} the two players (the bass token falls back to the treble instrument when the bass is set to same). The settings tokens record the Render tab: {reverb-type} the room, {reverb-amount} the two reverb sends as treble-bass (for example 50-20), {highpass} and {lowpass} the two filter values — useful when you're comparing test renders and want each file name to carry the exact settings it was made with. Everything else is kept as you typed it.

A new seed starts at {name}-{style}-{bpm}, so a seed called Chorus lift rendered as a pad at 80 BPM comes out as Chorus lift-pad-80. If the seed has no name, {name} falls back to the chord-derived name, so the file is never left starting with a dash. Because the template is stored unfilled and resolved fresh on each render, changing a setting — a new style, a slower tempo — renames the next render instead of overwriting the last one. Clear the field entirely and it goes back to the default template.

There is no output folder to choose: every render is written into the same folder as the seed's own .yams file. So save the seed first — the Render seed button stays disabled until you have (see Save, open and manage seeds) — and the audio and MIDI land right next to it.

See the seed before you render

On the Render tab, a small readout shows what the current form will produce before you render it: the number of chords (Chords), the total length (Duration) and the estimated file size (Est. size). It updates as you type, so a runaway seed — a long chart times many loops can run to minutes of audio — is visible up front instead of surfacing as a long freeze.

It's also a safety check. If the time signature can't be read, or the progression contains a chord the app doesn't recognise, the readout says so and the Render seed button stays disabled until you fix it — so you never start a render that was going to fail. While a render is running, the button shows its stage, first "Rendering audio…" and then "Saving…", so you can see it's working rather than frozen.

The files a render produces

Each render saves two files that share the output name: the audio seed (.mp3 or .wav) and a MIDI file (.mid) of the same progression. The result panel lists both, and clicking either one reveals it in your file manager. They land in the folder of the seed's .yams, right beside it.

They're for two different jobs. The audio is the seed you actually upload to the AI generator — it's the whole point of the app. The MIDI is a bonus for music software: open it in a DAW and you have the same progression as editable notes — it carries the harmony itself, complete even if you rendered with a player muted. If you only ever upload to Suno, the MIDI costs nothing sitting there. The .yams seed file in that same folder is a third thing, but it isn't produced by rendering — it's the document you saved to create the seed, covered next.

Save, open and manage seeds

A seed isn't just a one-off render — it's a small document you can save, name, reopen and keep several of open at once. Everything you type is captured in a .yams file, so a seed comes back exactly as it was, ready to re-render or tweak. Nothing is kept automatically between launches, so saving is how you hold on to a seed.

A seed is a `.yams` document

A seed lives as a .yams file — a small text document holding every field on the form: its name, the chords, the tempo, the meter, the style, both instruments and their mixer settings, the room and filters, the format, the output template, all of it. Saving one is what turns a throwaway render into a seed you can keep.

A figure you drew yourself is saved with it, in full — not as a reference to the style you started from, but as the notes themselves. That is deliberate: a seed should sound the same in a year even if the shipped figure it was drawn from is later improved. The style's name is kept beside it as a record of where the figure came from.

Seeds saved by older versions still open normally, and the name their renders used is kept as the output template, so reopening one and rendering it again produces the same file names as before.

The app opens with a single blank seed each time, and nothing you type is remembered between launches — to keep a seed, save it. Three buttons in the top bar manage this:

New seed — start a fresh blank seed; it opens in a new tab.
Save seed — write the current seed back to its .yams. The first save of a new seed asks where to put the file.
Save seed as… — save the current seed to a new .yams, leaving the original untouched — the way to fork a variant.

Saving is also the gate on rendering: a seed must have a .yams before Render seed works, because the audio is written into that file's folder.

Several seeds open at once

You can have several seeds open together, each in its own tab below the top bar — the same way a text editor holds several open files. Click a tab to switch to it; each tab keeps its own chords, settings and last render independently.

A tab shows the seed's file name, or Untitled for one you haven't saved yet. A small dot appears on a tab whose seed has unsaved edits, so you can see at a glance what still needs saving. The Close button on a tab closes it; the app never sits empty, so closing the last tab drops a fresh blank seed in its place.

Open a saved seed

Bringing an existing seed back into the app can happen three ways, and they all end the same — the seed opens in a tab you can render:

Load seed — the Load button in the top bar opens a file picker filtered to .yams seeds.
Drag and drop — drag a .yams file straight onto the app window and drop it. A soft overlay appears while you're dragging, to show it will be loaded.
Double-click — because the app registers the .yams file type, double-clicking a seed (or using Open with) launches the app with that seed loaded, or hands it to an already-running window.

Opening a file that's already in a tab brings that tab forward, refreshed from the file, rather than duplicating it. However it arrives, the whole form fills back in from the file — chords, tempo, style, instruments, everything — because a seed is defined entirely by the form. Reopen one and press Render to get the same audio again, or change a single setting — a different style, a slower tempo — and render a variant without retyping the progression.

Language and appearance

Open the settings dialog from the gear button in the top bar to change two things.

Language — the language the app's text is shown in. Your choice is remembered for next time.
Theme — switches between a light and a dark look. Pick whichever is easier on your eyes; it has no effect on the seeds you generate.

Questions and answers

How do I actually use the seed in Suno? Render the seed, then in Suno start a Cover (or Extend) and upload the audio file the app saved. Suno listens to the harmony in that audio and builds its track on your chords. The chord names you typed never go into Suno — only the sound does.

Can I paste in a chord sheet I already have? Usually yes. The app reads each line as a bar and ignores [Section] headings, blank lines and | bar marks, so most plain chord charts drop in almost unchanged. Check the preview's chord count and the "unrecognised" warning to catch anything it couldn't read.

I clicked a note and it disappeared instead of letting me change it. That's the drawing mode doing exactly what it says: a click on a note removes it. To adjust a note you already drew, switch the mode control to Note settings: click a note to select it, then set its flags — or Ctrl+click the note, which selects it without leaving drawing mode. See Placing notes, and the two click modes.

The style menu says Custom — did I break something? No. It means the figure being played is no longer the preset's, because you edited the grid or swapped one player's figure. Your version is saved with the seed and plays exactly as you left it. Picking a style from the menu again replaces it with that preset.

Why is there a MIDI file, and what's the .yams? The MIDI is a convenience for music software — the same progression as editable notes for a DAW. The .yams is the seed's own document — the file you save to keep a seed, and reopen (or double-click) later to re-render or tweak it. Neither is what you upload to Suno — generators can't hear a MIDI, and the .yams is just settings. Upload the audio; the other two are there if you want them.

Which style and instruments should I pick? When in doubt: Pad, one instrument (leave the bass on Same as instrument), and MP3 at 192 kbps. That states the harmony clearly without dominating the result. Then use Play to explore — styles, instruments and the mixer all change live, so your ear can decide before anything is rendered.

Play sounds fine but I want the bass quieter in the file — do I have to re-render to check? No. The mixer knobs are live: adjust the bass volume while it plays and you're hearing the same mix the render will write. What you hear is what the file gets.

A banner says a new version is available. That's the app telling you a newer release has been published. Use the download link to get it, or dismiss the banner — there's no automatic update, so nothing changes until you choose to install it.

Instrument credits

The sampled instruments come from freely-licensed recording sessions. Thank you to the people who recorded and shared them:

Author	Instruments
Sam Gossner / Versilian Studios (VSCO-2-CE)	Bassoon, Cellos, Chimes, Clarinet, Double Bass, Flute, Harp, Horn, Marimba I, Marimba II, Oboe, Old Trombone, Organ, Piccolo, Recorder, Timpani, Trombone, Trumpet, Tuba, Upright Piano ×2, Violas, Violin, Violins
FreePats project (incl. roberto@zenvoid.org, Mateusz Dąbrowski, Strix SoundFont Team)	Accordion, Bagpipe, Electric Guitar, Fingered and Picked Bass, Glass, Hang, Jaw Harp, Kalimba, Nylon Guitar, Ocarina, Synth Pads (New Age, Bowed, Sweep, Choir), Tenor Saxophone, Tubular Bells, Ukulele, Xylophone
Alexander Holm	Salamander Grand Piano
Zenph Studios / OLPC, bank by roberto@zenvoid.org	YDP Grand Piano
Signature Sounds	Choir Aaaah's
pjcohen	Celesta, Upright Bass
Tim Kahn	Rhodes Mark II
OldBassMan	Wurlitzer 200A
Project16	Rickenbacker Bass